Student success, including enrollment, degree completion, and improved retention, has been adversely affected due to the covid-19 pandemic across the higher academic landscape in the United States. It is quite a challenge to accurately predict what factors impact student retention in an academic institution most. This research aims to develop and empirically test a comprehensive list of factors contributing to student retention behavior using data mining techniques (DMT). This study examines retention behavior prediction from the dataset (pre-pandemic and post-pandemic) of 18,000+ students enrolled at an academic institution. Here, we deploy six different machine learning (ML) models (e.g., Logistic Regression (LR), Support Vector Machine (Linear Classifier), Support Vector Machine (Radial Basis Function), Decision Tree (DT), Random Forest (RF), K-Nearest Neighbors (KNN) to predict student retention. Synthetic Minority Oversampling Technique (SMOTE) is also being used to fix the imbalance class problem and improve the performance of data mining algorithms. Empirical results showed that completed credits, grade point average (GPA), college entrance age, and attempted credits were vital factors in student retention behavior. This research also highlighted that RF algorithms outperformed other DMTs' and achieved the highest accuracy (0.86) with SMOTE for retention prediction. The critical results discussed here should apply to similar higher academic institutions worldwide. Academic institutions would benefit from launching preventative measures to avoid dropouts and improve retention based on the retention metrics reported here.